A rapid review of complex function theory

T. PERUTZ

I have followed W. Rudin, Real and complex analysis (chapter 10) and H. Priestley, Introduction to
complex analysis.

Notational conventions
• C is the complex plane.
• Open discs: D(a; r) := {z ∈ C : |z − a| < r}.
• Closed discs: D̄(a; r) = {z ∈ C : |z − a| ≤ r}.
• Punctured open discs: D∗ (a; r) = {z ∈ C : 0 < |z − a| < r}.

1 Holomorphic functions
Definition 1.1 Let G be an open set in C. A function f : G → C is called holomorphic if, at every
point z ∈ G, the complex derivative
f (z + h) − f (z)
f 0 (z) = lim
h→0 h
exists as a complex number.

Write O(G) for the set of holomorphic functions on G.

1.1 Basic properties of holomorphic functions

• If G and H are open sets in C, and H ⊂ G, then the restriction to H of a holomorphic function
on G is again holomorphic; thus one has a restriction map O(G) → O(H).
• The sum, product, chain and quotient rules hold for holomorphic functions, just as for differen-
tiable functions of a real variable.
• Since sums and products of holomorphic functions are again holomorphic, O(G) is a commutative
ring. Since constant functions are holomorphic, there is a ring homomorphism C → O(G),
making O(G) a commutative, associative C-algebra.

1.2 The derivative as a linear map

If f is holomorphic, one has

(1) f (z + h) = f (z) + f 0 (z)h + εz (h), εz (h) ∈ o(h),
i.e., εz (h)/h → 0 as h → 0. Conversely, if f (z + h) = f (z) + g(z)h + εz (h), where εz (h) ∈ o(h) for all
z, then f is holomorphic with derivative g.
From (1) one sees that a holomorphic function is differentiable in the real sense, as a function G →
C = R2 —that is, there is an R-linear map (the derivative) Dz f : C → C such that f 0 (z + h) =
2 T. Perutz

f (z) + (Dz f )(h) + εz (h) where εz ∈ o(h). The derivative is (Dz f )(h) = f 0 (z)h. It is not only R-linear
but also C-linear.
Complex linearity of the derivative is equivalent to the Cauchy–Riemann equation
(2) Dz (ih) = iDz (h)
In terms of real and imaginary parts, a C-linear map C → C is represented by a real matrix of form
 
a b
.
−b a
On the other hand, writing z = x + yi, h = h1 + h2 i, and f = u + vi (all variables on the right real) one
has     
h1 ux uy h1
(Df z) =
h2 vx vy h2
so we see that the Cauchy–Riemann equation can be written in the familiar form
ux = vy , uy = −vx .
Conversely, a function f : G → C that is R-differentiable and satisfies the Cauchy–Riemann equation
is holomorphic, since the derivative Dz f must be multiplication by a complex number f 0 (z).
While differentiability of a function f : G → R2 is a regularity (or smoothness) condition, that is not
true of holomorphy. For f to be holomorphic it must also satisfy a PDE, and that leads to properties of
holomorphic functions that are drastically more special than those of real differentiable functions.

1.3 Power series

P∞
For a complex power series − c)n , centered at c, we define the radius of convergence
n=0 an (z
∞
( )
X
R = sup |z − c| : an (z − c)n converges ∈ [0, ∞].
n=0
P∞ n
Theorem 1.2 Let n=0 an (z − c) be a power series with radius of convergence R. Then
(1) R−1 = lim sup |an |1/n .
The sum ∞ n
P
(2) n=0 an (z − c) converges for z ∈ D(c; R), to a function f (z). The convergence is
absolute.
(3) The convergence to f is uniform on compact subsets of D(c; R).
(4) The series ∞ n−1 also has radius of convergence R, and so converges on D(c; R)
P
n=1 nan (z − c)
to a function g.
(5) f ∈ O(D(c; R)), and f 0 = g.

The proof is basically a comparison with geometric series.

Proof (1 and 2) It follows from the root test that if lim sup |an |1/n |z − c| > 1 then the series diverges,
while if lim sup |an |1/n |z − c| < 1 then it converges absolutely (hence converges).
(3) It suffices to prove uniform convergence on D(c; ρ) where ρ < R. On this disc one has |an (z−c)n | ≤
Mn , where Mn = |an |ρn . But
P
Mn < ∞, and so uniform convergence follows from the Weierstrass
M-test.
A rapid review of complex function theory 3

(4) Apply (1), noting that lim sup{(n + 1)1/n |an+1 |1/n } = lim sup |an |1/n .
(5) We may assume c = 0. Absolute convergence of f and g validates the formula
f (z) − f (w) X  zn − wn 
n−1
− g(w) = an − nw .
z−w z−w
n≥1

for distinct z, w ∈ D(c; R). One has

zn − wn
 n
z − wn

n−1 ∂ n−1
− nw = (z − w) − nw = (z − w)(zn−2 + 2zn−3 w + · · · + (n − 1)wn−1 )
z−w ∂w z−w
(the sum on the right is understood to be zero if n = 1). Thus if |z| and |w| are less than ρ < R, one
has
zn − wn 1
− nwn−1 ≤ n(n − 1)ρn−2 |z − w|
z−w 2
and
∞
!
f (z) − f (w) X 1 n−2 1
− g(w) ≤ n(n − 1)ρ |z − w| = |z − w|,
z−w 2 (1 − ρ)3
n=2

which goes to 0 as z → w.

2 Cauchy’s theorem

2.1 Path integrals

Definition 2.1 If G ⊂ C is an open set, a smooth path in G is a compact non-empty interval [a, b]
and a C1 map γ : [a, b] → G. A path in G is a compact non-empty interval [a, b] and a continuous
map γ[a, b] → G such that [a, b] is the union of finitely many closed subintervals J such that γ|J is a
smooth path. A loop is a path such that γ(a) = γ(b).

For a smooth path γ : [a, b] → G, and for a continuous function f : G → C, we define the integral of f
along γ to be
Z Z b
f (z) dz := f ◦ γ(t) · γ 0 (t) dt.
γ a
R R
We also write γ f for γ f (z) dz.
This integral behaves in the expected Rfashion under reparametrization, and is additive under joining
paths. For a general path γ , we define γ f (z) dz by summing integrals over subintervals on which γ is
smooth.
The fundamental theorem of calculus holds: if F ∈ O(G) then γ F 0 (z)dz = F(γ(b)) − F(γ(a)). In
R

particular:

Proposition 2.2 If γ : [a, b] → G is a loop and F ∈ O(G) then F 0 (z)dz = 0.

R
γ
4 T. Perutz

2.2 Proving Cauchy’s theorem

Given three points v1 , v2 and v3 in C, we define the triangle T = T(v1 , v2 , v3 ) to be their convex
hull. Its boundary ∂T is the loop formed as the join of the three directed intervals [v1 , v2 ], [v2 , v3 ] and
[v3 , v1 ], each parametrized with unit speed.

Lemma 2.3 (Cauchy’s theorem for a triangle)

R Let f ∈ O(G); let v1 , v2 and v3 be points of G whose
convex hull T is contained in G. Then ∂T f = 0.

Proof Let v12 = 12 (v1 + v2 ); similarly define v23 and v31 . We subdivide the triangle T into four
congruent triangles:
T 1 = T(v1 , v12 , v31 ), T 2 = T(v2 , v23 , v12 ), T 3 = T(v3 , v31 , v23 ), T 4 = T(v12 , v23 , v31 ).
Then ∂T = 4k=1 ∂T k , because of cancelations. Let T1 be a triangle T k which maximizes ∂T k f .
R P R R

Thus Z Z
f ≤4 f .
∂T ∂T1
Similarly subdivide T1 , and iterate the process to obtain a nested sequence of triangles TN such that
Z Z
N
f ≤4 f .
∂T ∂TN
Let z∗ be the unique point in TN . One has f (z) = f (z∗ )+f 0 (z∗ )(z−z∗ )+ε(z), where ε(z)/|z−z∗ | → 0
T
as z → z∗ . Now Z
f (z∗ ) + f 0 (z∗ )(z − z∗ ) dz = 0


∂TN
by Proposition 2.2, and so Z Z
f = ε.
∂TN ∂TN
Thus Z Z
f ≤ 4N ε ≤ 4N · 2−N length (γ) sup |ε(z)|.
∂T ∂TN z∈∂TN

For any δ > 0, we can find a disc centered on on which |ε(z)| ≤ |z − z∗ |δ . Thus if N is sufficiently
z∗
large one has supz∈∂TN |ε(z)| ≤ 2−N δ , whence
Z
f ≤ length (γ) · 2N · sup |ε(z)| ≤ length(γ) · δ.
∂T z∈∂TN

Theorem 2.4 (Cauchy’s theorem for R a star-domain) Let G be a star-domain, and f ∈ O(G). Then
for any closed path γ in G one has γ f = 0.

Proof Let z∗ be a star-point; thus for all z ∈ G, the interval [z∗ , z] lies in G. Define F(z) =
0
R
[z,z∗ ] f (z) dz. It suffices to show that F = f .
Take z ∈ G, and let δ > 0 be small enough that D(z; δ) ⊂ G. By Cauchy’s theorem for a triangle one
has, when |h| < δ , Z
F(z + h) − F(z) = f.
[z,z+h]
A rapid review of complex function theory 5

Applying this formula to the expression f (z + h) = f (z) + f 0 (z)h + εz (h), one obtains
Z 1
1 0 2
F(z + h) = F(z) + f (z)h + f (z)h + hεz (th) dt,
2 0

from which one obtains F 0 = f .

Corollary 2.5 (deformation theorem) Let G beRan openR set, and let γ0 and γ1 be closed paths that
are homotopic in G (through closed paths). Then γ0 f = γ1 f for all f ∈ O(G).

One takes the homotopy to be piecewise C1 in an appropriate sense. The proof is to subdivide the
homotopy into many small homotopies whose images lie inside discs in G, and then to apply Cauchy’s
theorem for a star-domain to each of those.

Corollary 2.6 (Cauchy’s theorem for simply connected regions) Let G be a simply connected open
set in C, and let γ be a closed path in G. Then γ f = 0 for all f ∈ O(G).
R

Corollary 2.7 Let G be a simply connected open set in C. Then every holomorphic function on G
possesses an antiderivative on G.
R
One constructs the antiderivative F as F(z) = γz f , where γz is an arbitrary path connected a basepoint
z0 ∈ G to z.

2.3 From Green’s theorem to Cauchy’s theorem.

Recall that Green’s theorem states that if D is a compact set in the plane whose boundary ∂D is a
smooth curve then, for the counterclockwise orientation of ∂D, one has
ZZ Z
(Gx − Fy )dx dy = F dx + G dy
D ∂D

provided that F and G are C1 functions. If f = u + vi is holomorphic and C1 , one gets

Z Z ZZ
Re f (z)dz = u dx + v dy = (vx − uy ) dx dy.
∂D ∂D D
R R
But vx −uy = 0 by
R the Cauchy–Riemann equations, so Re ∂D f (z)dz = 0. Similarly Im ∂D f (z)dz = 0.
This shows that ∂D f (z)dz = 0, a form of Cauchy’s theorem. (This argument is essentially the one that
Cauchy himself used.)
The drawback of this approach is that it requires one to assume f is C1 . It is true that every holomorphic
function is C1 , but the usual proof is via Cauchy’s theorem. To avoid circularity, if one takes this
approach one must either assume that one’s functions are holomorphic and C1 (which is aesthetically
displeasing, though not really problematic in practice), or find an alternative proof that holomorphic
functions are C1 . There is such an alternative, known as Weyl’s lemma (or as elliptic regularity), and
we will prove it (for other reasons) in this course.
6 T. Perutz

3 Power series representations of holomorphic functions

3.1 Integral formulae

Theorem 3.1 (Cauchy’s integral formula) Let G be an open set containing the disc D(z; R) and take
f ∈ O(G). Then, for r ∈ (0, R), one has
Z
1 f (w)
f (z) = dw.
2πi C(z;r) w − z

Here C(z; r) denotes the circular path [0, 2π] 3 t 7→ z + reit .

Proof One has

f (w) − f (z)
Z Z
1 f (w) 1
dw − f (z) = dw.
2πi C(z;r) w−z 2πi C(z;r) w−z
Since the integrand in the latter expression is holomorphic on the punctured disc D∗ (z; r), one has by
the deformation theorem that the integral does not depend on r ∈ (0, R). But we can find a disc centered
at z on which
f (w) − f (z)
− f 0 (z) ≤ |w − z|.
w−z
1
R f (w) 0 2
Taking r smaller than the radius of this disc, we find that 2πi C(z;r) w−z dw − f (z) ≤ |f (z)|r + r , and
the right-hand side goes to zero as r → 0.

Theorem 3.2 (holomorphic functions are analytic) Let f ∈ O(D(c; R). Then one has f (z) =
n
P
n≥0 an (z − c) on D(c; R), where for any r ∈ (0, R) one has
Z
1 f (z)
an = dz
2πi C(c;r) (z − c)n+1

Proof We use Cauchy’s integral formula and the deformation theorem. Take δ > 0 small, and then
take r ∈ (|z − c|, R). Then
Z
1 f (w)
f (z) = dw (for small δ > 0)
2πi C(z;δ) w − z
Z
1 f (w)
= dw
2πi C(c;r) w − z
Z !
1 f (w) 1
= z−c dw
2πi C(c;r) w − c 1 − w−c
f (w) X z − c n
Z  
1
= dw
2πi C(c;r) w − c w−c
n≥0
X 1 Z f (w)

= dw (z − c)n .
2πi C(c;r) (w − c)n+1
n≥0
In the formula for an , we know by the deformation theorem that any r ∈ (0, R) will give the same
integral.
A rapid review of complex function theory 7

Corollary 3.3 (Cauchy’s formula for derivatives) Holomorphic functions are C∞ . For f ∈ O(G)
one has Z
(n) n! f (w)
f (z) = dw
2πi C(z;r) (w − z)n+1

for small r > 0.

3.2 Discrete zeros

Theorem 3.4 Let f ∈ O(G), where G ⊂ C is an open set. If f is not the zero-function then f −1 (0) is
discrete in G.

Proof Suppose f (c) = 0, and let f (z) = ∞ n

P
n=1 an (z−c) be the Taylor series representation of f near c.
If f is not the zero-function then there is some non-vanishing coefficient an . Let m be
Pthe smallest integer
such that am is non-vanishing. Then we have f (z) = (z − a)m g(z), where g(z) = ∞ n=m an (z − c)
n−m .

The function g isPholomorphic near a, since it is a power series with positive radius of convergence,
equal to that for ∞ n
n=1 an (z − c) , and g(a) = am 6= 0. Thus g is non-vanishing near a, and hence f is
non-vanishing in a punctured neighborhood of a.

The integer m that appears in the proof is called the multiplicity of the zero, and is denoted by mult(f ; c).

3.3 Liouville’s theorem

Theorem 3.5 (Liouville’s theorem) If f ∈ O(C) is bounded then f is constant.

Proof Say |f | ≤ M . We have

Z
1 f (w)
f 0 (z) = dw,
πi C(0;r) (w − z)2

for any r > |z|. Taking r > 2|z|, we have |w − z| ≥ |w| − |z| ≥ r/2 for w ∈ C(0; r). Thus
|f 0 (z)| ≤ π1 · 2πr · 4Mr−2 = 8Mr−1 , whence f 0 (z) = 0. Since z is arbitrary, f is constant.

4 Morera’s theorem and its consequences

R
Theorem 4.1 (Morera’s theorem) If f is a continuous function on the open set G ⊂ C, and ∂T f =0
for all triangles T ⊂ G, then f ∈ O(G).

R
Proof On a disc D(c; r) ⊂ G, we can define an antiderivative F for f by the formula F(z) = [c,z] f
(the proof that F 0 = f is as in the proof of Cauchy’s theorem for a triangle). Since F is holomorphic,
so is its derivative f .
8 T. Perutz

4.1 Removable singularities

Corollary 4.2 Let f be holomorphic on an open set G containing a punctured disc D∗ (a; r) on which
|f | is bounded. Then f extends to a holomorphic function on G ∪ {a}.

In this situation, we say that f has a removable singularity at a.

Proof Let g(z) = (z − a)f (z). Then g ∈ O(G), and g(z) → 0 as z → a. Hence g can be continuously
extended to G ∪ {a} by setting g(a) = 0. One R now checks that Morera’s theorem applies to g on
G ∪ {a}: take a triangle T ⊂ G. If a ∈ / T then ∂T g = 0 by Cauchy’s theorem. If a ∈ ∂T then we can
replace
R T by a closed curve γ not containing a at the cost of an arbitrarily small change in the integral;
thus ∂T g = 0. If a ∈ int(T) then by the deformation theorem we can replace T by an arbitrarily small
triangle T 0 whose interior contains T . But ∂T 0 g can be made arbitrarily small, hence ∂T g = 0.
R R

By Morera, g is holomorphic over a. But g(a) = 0, from which it follows that (z − a)−1 g(z) is again
holomorphic over a. This is the sought extension of f .

4.2 Limits of sequences of holomorphic functions

Theorem 4.3 Let (fn )∞n=0 be a sequence of holomorphic functions on an open set G converging,
uniformly on compact subsets of G, to a limit f . Then f ∈ O(G), and fn0 (z) → f 0 (z).

Proof We may work in an open set U contained in a compact subset of G, so thatR convergence R is
uniform. Uniform convergence implies both that the limit f is continuous, and that γ fn → γ f for
R R
paths γ in U . Taking γ to be an arbitrary triangle, we have γ fn = 0 by Cauchy, so γ f = 0, so f
is holomorphic by Morera. That fn0 (z) → f 0 (z) follows from Cauchy’s integral formula for derivatives.

In practice, one often wants to combine this theorem with the Weierstrass M-test, which states that if
one has P , and if there are constants Mn ≥ 0 such that |fn | ≤ Mn on
a sequence of functions (fn ) on UP
U , and Mn < ∞, then the sum f (z) = ∞ n=0 fn (z) converges uniformly on U :

Corollary 4.4 Let (fn )∞

n=0 be a sequence of holomorphicP functions on an open set UP
. Assume that
there are constants Mn ≥ 0 such that |fP
n | ≤ Mn on U , and Mn < ∞. Then the sum fn converges
to a holomorphic function f , and f 0 = fn0 .

5 Isolated singularities

Theorem 5.1 Consider the annulus A = A(c; r1 , r2 ) = {z ∈ C : r1 < |z−c| < r2 }, and take f ∈ O(A).
Then one can write X X
f (z) = an (z − c)n + an (z − c)n ,
n<0 n≥0

1
R f (z)
where an = 2πi C(c;r) (z−c)n+1 dz.
A rapid review of complex function theory 9

Sketch of proof Fix z ∈ A. Choose r10 and r20 so that r1 < r10 < |z − c| < r20 < r2 . Then one has
Z Z
1 f (w) 1 f (w)
f (z) = dw − dw,
2πi C(c;r10 ) w − z 2πi C(c;r20 ) w − z
as one sees by splitting the closed annulus bounded by C(c; r10 ) and C(c; r20 ) into two contractible
regions, the first containing z, the second not, and applying Cauchy’s integral formula and Cauchy’s
theorem to these respective regions. One then develops binomial expansions for the two integrands,
like in the proof that holomorphic functions are analytic.
If f ∈ O(G), where G contains a punctured disc centered at c, the residue of f at c, denoted res{f ; c},
is the coefficient a−1 in the Laurent expansion. ThusZ
1
res{f ; c} = f (z) dz
2πi C(c;r)
for any small r > 0.
Theorem 5.2 (Residue theorem) Let D be a compact, simply connected region bounded by a loop
∂D. Let a1 , . . . , ak be interior points of D, ∗ = int D \ {a , . . . , a }, and take f ∈ O(D∗ ). Then
Z let D X 1 k
1
f = res{f ; aj }.
2πi ∂D j

Sketch proof The argument is to consider a null-homotopy {γt } of the loop γ0 = ∂D. As t increase
towards 1,R the aj are expelled from the region bounded by γt . Using the deformation theorem, one
finds that ∂D f is the sum of the integrals of f around small loops encircling the aj , i.e., 2πi times the
sum of the residues.

5.1 Counting zeros

Theorem 5.3 (Argument principle) Let D be a compact, simply connected region bounded by a loop
∂D. Assume that f is holomorphic on a neighborhood of D, and non-zero on ∂D. Then
f 0 (z)
Z
1 X
(3) dz = mult (f ; z).
2πi ∂D f (z)
z∈D: f (z)=0
f 0 (z)
Z
1 X
(4) z· dz = f (z) · mult (f ; z).
2πi ∂D f (z)
z∈D: f (z)=0

Proof The logarithmic derivative f 0 /f is holomorphic except at the discrete set f −1 (0). Enumerate the
R f 0 (z)
finite set f −1 (0) ∩ D as (a1 , . . . , ak ). By the residue theorem, 2πi 1
∂D f (z) dz is the sum of the residues
of f 0 /f at the points aj . Thus, to prove the first formula, it suffices to show that if f has a zero of
multiplicity m at a point a then res {f 0 /f , a} = m. We may write f (z) = (z − a)m g(z), where g is
holomorphic near a and g(a) 6= 0. Then
f 0 (z) m g0 (z)
= + ,
f (z) z−a g(z)
which makes it clear that the residue is m. The proof of the second formula works in the same way.

Corollary 5.4 (Open mapping theorem) Let G be a connected open set and f ∈ O(G) non-constant.
Then f is an open mapping, i.e., it maps open sets to open sets.